Research map / Label-efficient and robust learning / Recognition and long-tail
SOLAR: Second-Order Loss and Attention for Image Retrieval
Tony Ng, Vassileios Balntas, Yurun Tian, Krystian Mikolajczyk
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add second-order similarity (SOS) loss to GeM baseline (ablation) | ResNet101-GeM · ROxf Hard | mAP: 44.3 -> 44.9 | Improved |
| add SOA block after conv4_x (ablation) | ResNet101-GeM · ROxf Hard | mAP: 44.3 -> 45.7 | Improved |
| add SOA block after conv5_x (ablation) | ResNet101-GeM · ROxf Hard | mAP: 44.3 -> 45.9 | Improved |
| add consecutive SOA blocks after conv4_x and conv5_x (ablation) | ResNet101-GeM · ROxf Hard | mAP: 44.3 -> 46.9 | Improved |
| add SOS loss on top of SOA blocks (full SOLAR) (ablation) | ResNet101+SOA · ROxf Hard | mAP: 46.9 -> 47.9 | Improved |
| add SOS loss on top of SOA blocks (full SOLAR) (ablation) | ResNet101+SOA · RPar Hard | mAP: 63.7 -> 64.5 | Improved |
| add SOLAR instead of GeM baseline on 1M distractors | ResNet101-GeM · RPar+R1M Hard | mAP: 29.8 -> 33.4 | Improved |
| add SOLAR instead of GeM baseline | ResNet101-GeM · ROxf Medium | mAP: 67.3 -> 69.9 | Improved |
| add SOA block after layer 3 of SOSNet (ablation) | SOSNet+ · UBC Patches | FPR@95 mean: 1.138 -> 1.065 | Improved |
| add SOA block after low-resolution layer 6 of SOSNet (ablation) | SOSNet+ · UBC Patches | FPR@95 mean: 1.138 -> 1.380 | Hurt |
| insert SOA at consecutive layers (3,4) rather than non-consecutive (3,5) (ablation) | SOSNet+ · HPatches Patch Retrieval | mAP: 0.17% | Improved |
| insert SOA at consecutive layers (4,5) rather than (4,6) (ablation) | SOSNet+ · HPatches Patch Retrieval | mAP: 0.45% | Improved |
Closest papers by meaning
- Deep Image Retrieval: Learning Global Representations for Image Search (ECCV 2016 · similarity 0.85)
- Learning Super-Features for Image Retrieval (ICLR 2022 · similarity 0.85)
- End-to-End Learning of Deep Visual Representations for Image Retrieval (IJCV 2017 · similarity 0.85)
- SORT: Second-Order Response Transform for Visual Recognition (ICCV 2017 · similarity 0.85)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.83)
- How to Make Cross Encoder a Good Teacher for Efficient Image-Text Retrieval? (CVPR 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
